Note : Les descriptions sont présentées dans la langue officielle dans laquelle elles ont été soumises.
CA 02475183 2004-07-20
A METHOD TO REDUCE ACOUSTIC COUPLING IN AUDIO CONFERENCING
SYSTEMS
BACKGROUND OF THE INVENTION
1. Field of the Invention
[0001] The present invention is directed to audio conferencing systems, and
more particularly to
a method of reducing acoustic coupling and howling effects in a full duplex
audio conferencing
system using beamforming.
2. Description of the Related Art
[0002] Beamforming technology (also known as microphone array technology) can
be used for
spatial directivity of sound pickup in a conferencing system to enhance the
quality of near-end
speech. For such systems that perform full-duplex communications, acoustic
echo cancellation
presents numerous challenges.
[0003] One problem that is well known ii ~ the art relates to combining
acoustic echo cancellation
with beamforming (see M. Brandstein and D. Ward, "Microphone Arrays. Signal
Processing
Techniques and Applications°. Springer Verlag, 2001, and H. Buchner, W.
Herbordt, W.
Kellermann, °An Efficient Combination of Multi-Channel Acoustic Echo
Cancellation With a
Beamforming Microphone Array', Proc. Int. Workshop on Hands-Free Speech
Communication
(HSC), pp. 55-58, Kyoto, Japan, April, 2001 ). One approach is to perform
acoustic echo
cancellation on all the microphone signals in parallel, which is
computationally intensive. A
second approach is to perform acoustic echo cancellation on the spatially
filtered signal at the
output of the beamformer.
[0004] In the approach where acoustic echo cancellation is performed on the
spatially filtered
signal at the output of the beamformer, it is clearly desirable that the
beamformer presents good
loudspeaker-to-beam coupling characteristics in order to minimize the amount
of echo that has
to be cancelled. At the same time, it is also desirable that the beamformer
has good directivity in
its look direction to provide enhanced speech quality based on spatial
directivity. In practice,
however, it is~impossible to meet both of these requirements with the same
beamformer design
as there typically is a trade off between the beamformer's directivity and
response to the
loudspeaker signal, as discussed below.
[0005] A number of design requirements must be met to ensure good coupling
characteristics.
Firstly, the beamformer must be robust to uncorrelated perturbations of the
microphone signals.
The reason for this is that in a conference system, the physical coupling
between the
loudspeaker and the microphones is subject to variations caused by the
loudspeaker and other
CA 02475183 2004-07-20
2
structural vibrations or leakage. Therefore, a high sensitivity of the
beamformer to such
variations may result in excessive acoustic coupling between the loudspeaker
and the output of
the beam. This, in turn, affects the performance of the acoustic echo
cancellation, resulting in
echo bursts as well as howling effects due to positive loop gain. It should be
noted that the
sensitivity problem applies to both structures described above for combining
acoustic echo
cancellation with beamforming. indeed, even if acoustic echo cancellation is
performed on the
microphone signals in parallel, the residual echo signals are not free of
amplitude and phase
variations due to structural coupling (vibrations and acoustical leaks). Many
of these variations
will remain on the microphone signals after the individual echo cancellation.
When combined in
the beamformer they may result in a large error signal if the beamformer is
sensitive to the
variations described above. The second design requirement that must be met to
ensure good
coupling characteristics is that the beamformer should also provide strong
attenuation of the
loudspeaker signal (which can be seen as an interference signal) in order to
achieve better echo
cancellation performance in the conferencing system.
BRIEF DESCRIPTION OF THE DRAWIMGS
[0006] A further description of the prior art and of a preferred embodiment of
the invention is set
forth in detail below with reference to the following drawings, in which:
[0007] Figure 1 shows a series of polar beampattem plots for a circular array
of six
microphones located around a diffracting structure with a loudspeaker located
at the center,
illustrating the effect of regularization in accordance with the prior art.
[0008] Figure 2 is a graph showing the coupling response to the loudspeaker
signal with
random uncorrelated perturbations, for the circular array of six microphones
located around a
diffracting structure with a loudspeaker located at the center, in accordance
with the prior art.
[0009] Figure 3 shows a series of polar beampattern plots for a circular array
of six
microphones located around a diffracting structure with a loudspeaker located
at the center after
placing a null in the direction of an interference signal, in accordance with
the prior art.
[0010] Figure 4 is a graph showing the coupling response to the perturbed
loudspeaker signal
after pulling, for the circular array of six microphones located around a
diffracting structure with
a loudspeaker located at the center, in accordance with the prior art.
[0011] Figure 5 is a flowchart showing steps in the method according to the
present invention
for reducing audio coupling in an audio conferencing system.
CA 02475183 2004-07-20
3
DISCUSSION OF THE BACKGOUND ART
[0012] It is well known in the art (see Branstein et al, supra, chapter 2)
that there is a trade-off
between the directivity towards a talker and the robustness of the beamformer
to amplitude and
phase variations of the signals received by the microphones. In other words,
very directive
beamformers (designed with such methods as Minimum Variance Distortionless
Response or
other superdirective variants) are characterized by a high sensitivity to
uncorrelated noise in the
microphone signals. Figure 1 illustrates this fact with a circular array of 6
microphones located
around a diffracting structure with a loudspeaker located in the middle. The
thick dashed line
represents the beampattern of a standard minimum-variance distortionless
(MVDR) beamformer
with no special requirements for robustness whereas the thin line represents
an MVDR
beamformer designed for robustness (using a strong regularization factor as
explained below). It
can be seen that the beamformer designed for robustness presents a less
directive
beampattern. Figure 2 shows the effect of the robustness on the coupling
response when
random uncorrelated perturbations (of a maximum of 5dB in magnitude and 20
degrees in
phase in this particular case) are added to the signals received at the
microphones in response
to the loudspeaker signal. The line definitions ere the same as for Figure 1.
It can be seen that
the more directive and less robust beamformer shows a very strong perturbed
coupling
response that would present serious challenges for the acoustic echo canceller
in a practical
conference system.
[0013 The general problem of achieving directivity and robustness is well-
known in the art and
several solutions have been proposed. All of them achieve a compromise between
directivity
and robustness, by introducing a regularization factor in the design of the
beamformer (see for
instance E.N.Gilbert and S.P.Morgan, "Optimum design of directive antenna
arrays subject to
random variations", Bell Syst. Tech. J., pp 637-663, May 1955). This
regularization factor can
vary between 0 and +~. When given the value 0, the design results in a pure,
superdirective
beamformer (with high sensitivity). When given a value that tends to +~c, the
design then yields
the so-called conventional beamformer (also referred to as a delay-and-sum
beamformer, or the
Barlett beamformer (see Brandstein et al, supra, chapter 2), which is very
robust but not very
directive. Any value of the regularization factor between these two extremes
represents a
compromise between directivity and robustness. The process to find the best
compromise for a
particular device and a particular application is sometimes a. trial-and-error
process as in
M.Doerbecker, "Mehrkanalige Signalverarbeitung zur verbesserung akustisch
gestorter
CA 02475183 2004-07-20
4
sprachsignale am beispiel elektronischer hdrhilfen", PhD thesis, Dept of
Telecommunications,
University of TH Aachen, Verlag der Augustinus Buchhandlung, Aachen, Germany,
Aug 1998. it
is sometimes an iterative process as in J.M.Kates and M.R.Weiss, "A comparison
of hearing-aid
array-processing techniques", J. Acoustic. Soc. Amer., vol. 99, pp 3138-3148,
May 1996. All
these solutions achieve a trade off between directivity and coupling of the
beamformer thus
compromising the overall performance of the conferencing system.
(0014j The second design requirement mentioned above may be referred to as
interference
nuiling. It can be achieved with standard linear-constraint techniques as
described in Van Veen
and Buckley, °Beamforming: a versatile approach to spatial filtering",
IEEE ASSP Magazine,
April 1998. However, it is known in the art that placing a null (or strong
attenuation) in the
direction of an interference signal may also significantly affect the
beamformer directivity in
other directions because it reduces the number of degrees of freedom available
to form a beam
in its look direction. Figure 3 illustrates this fact with the same array as
discussed above in
connection with Figures 1 and 2. The thin tine represents a beampattern of a
standard MVDR
beamformer whereas the thick dashed line represents the MVDR beamformer with
an additional
linear constraint designed to place a null in the direction of the loudspeaker
signal. Both
beamformers have been designed with the same regularization factor. It can be
seen that the
interference pulling affects the directivity in a significant manner. Figure 4
illustrates the effect of
this interference pulling on the coupling response when random uncorrelated
perturbations (of a
maximum of 5dB in magnitude and 20 degrees in phase as in Figure 2) are added
to the signals
received at the microphones in response to the loudspeaker signal. The line
definitions are the
same as for Figure 3. It can be seen that the perturbed coupling response is
indeed reduced
over most of the frequency range.
SUMMARY OF THE INVENTION
[0015] The inventors have realized that for the particular application of
audio conferencing, the
beamformer does not require good coupling characteristics and good near-end
directivity at the
same time. Moreover, the conferencing device has internal knowledge of which
beamformer
feature is more important at any given point in time (depending on the
dynamics of the
conversation: far-end speech, near-end speech, etc). More specifically, the
inventors have
recognized that an audio conferencing system only requires spatial directivity
in the presence of
near-end speech, and it only requires good coupling characteristics in the
presence of far-end
speech. Therefore, the key technical aspect of the present invention is to use
two different
CA 02475183 2004-07-20
beamformers: one with good coupling characteristics for far-end-speech
(including double-talk)
and one with good directivity towards the near-end speaker, for near-end
speech. Double-
silence can be handled with one or the other scheme the choice between which
is an
implementation decision, based on the specific characteristics of the
particular device.
[0016] According to the foregoing, there is no need to compromise between
directivity and
coupling characteristics in the same beamformer design. The resulting system
provides
enhanced near-end speech quality (reverberations and local interference are
suppressed by the
"near-end" directional beamformer) as well as enhanced full-duplex operation
(loudspeaker
feedback is reduced by the "far-end" beamformer that is used in the presence
of far-end signal
or double-talk).
[0017] The present invention is to be distinguished from half duplex echo
cancellation, wherein
the transmit channel is closed during far-end speech activity and partially
closed during double-
silence. In the present invention, both sound pickup schemes can be designed
to provide a
similar gain and frequency response towards the near-end talker. The resulting
audio
conferencing system therefore provides true full-duplex communication, and
switching between
the two sound-pickup schemes is much less noticeable for the far-end listener
than it is with a
half duplex system. Furthermore, the full-duplex conferencing system benefits
from maximal
directivity when required (i.e. during near-end speech and possibly double-
silence) as well as
reduced loudspeaker-to-beam coupling performance when required (i.e. during
far-end speech,
double-talk and possibly double-silence).
[0018] The present invention relies on a good mechanism to detect the presence
of near-end
speech and/or far-end speech, This mechanism is not part of the present
invention but would be
well known to a person of ordinary skill in the art. In general, such a
mechanism is available in
conventional full-duplex acoustic echo cancellation algorithms that operate
based on reliable
decision logic regarding silence, single-talk and double-talk situations.
Examples of such
standalone mechanisms are set forth in U.S. Patent 6453041 (Voice activity
detection system
and method, Erol Eryilmaz), and U.S.Patent 5963901 (Method and device for
voice activity
detection in a communication device, Vahatalo et al).
[0019] DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENT
[0020] Figure 5 is a flowchart of the method steps according to the present
invention. Upon
CA 02475183 2004-07-20
6
receipt of a new sample or block of samples, the full-duplex acoustic echo
cancellation control
logic detects between near-end speech and far-end speech and outputs the
results of its
detection (i.e. far-end speech detected, near-end speech detected or near-end
speech not
detected). In the event that far-end speech is detected, the far-end speech
sound-pickup
scheme is used. That is, a beamformer is switched in to process the signal
samples with good
coupling characteristics for far-end-speech, double-silence and double-talk.
In the event that
near-end speech is detected, the near-end speech sound-pickup scheme is used.
That is, a
beamformer is switched in to process the signal samples with good directivity
towards the near-
end speaker. Similarly, if neither far-end nor near-end speech is detected,
then the far-end
speech sound-pickup scheme is used. As mentioned above, an alternative
implementation is to
use the near-end speech sound-pickup scheme for double-silence as well as for
near-end
speech.
[0021] Preferably, switching between the two sound pickup schemes is gradual
rather than
instantaneous. In particular, in the case where double-silence is handled by
the "far-end speech"
sound pickup scheme, when near-end speech starts or resumes in the near-end
location, the
system preferably switches gradually from the "far-end speech" sound-pickup
scheme to the
"near-end speech" sound pickup scheme. This results in a gradual audio focus
towards the
near-end talker. The speed of the transitions from one beamformer to the other
may depend on
the properties of the two sound pickup schemes used, as will be apparent to a
person of
ordinary skill in the art.
[0022] The design of the two sound-pickup schemes utilize standard beamforming
techniques
available in the literature (see Brandstein et ai, supra). Generally, the
design requirements are
as follows. The "far-end" beam needs good coupling response to the loudspeaker
signal (e.g
standard linear-constraint techniques combined with standard regularisation
techniques to
ensure good robustness). The "near-end" beam needs good spatial directivity
towards its look
direction in order to deliver enhanced speech quality to the far-end party. To
minimize transition
effects, both beams should present similar gain and frequency response in the
look direction of
interest.
[0023] Alternatives and variations of the invention are possible. For example,
the two sound-
pickup schemes do not specifically need to be beamformers, but can be any
sound pickup
schemes that provide spatial directivity in one case and coupling performance
in the other case,
while both offering full gain towards the near-end talker.
CA 02475183 2004-07-20
7
[0024] Another variation is to use a different "far-end° sound pickup
scheme for each look
direction instead of a single "far-end" beam common to all directions. The
"far-end" sound
pickup schemes would then have to present good coupling characteristics and
some directivity
towards their look directions. This approach solves the problem of compromise
between
coupling and directivity, but it does not solve the problem of having to deal
with acoustic echo
cancellation on a signal with time varying echo path, as discussed above.
[0025] The many features and advantages of the invention are apparent from the
detailed
specification and, thus, it is intended by the appended claims to cover all
such features and
advantages of the invention that fall within the sphere and scope of the
invention. Further, since
numerous modifications and changes will readily occur to those skilled in the
art, it is not desired
to limit the invention to the exact construction and operation illustrated and
described, and
accordingly ali suitable modifications and equivalents may be resorted to,
falling within the
scope of the invention.
